Section 9: Modern CPU Designs (Intel vs AMD)

The Big Picture: Same Language, Different Engines

Intel and AMD both build CPUs that run the same x86/x64 instructions (Section 8). A program built for one runs unchanged on the other. It is like two car makers who both build cars you drive with a steering wheel and pedals. What differs is the engine under the bonnet: how the internal hardware is engineered to run those instructions.

9.1 Intel Core Architecture

Modern Intel desktop and laptop CPUs (12th generation "Alder Lake" onward) use a hybrid architecture. That means two different core designs sit on one chip:

Intel Core i7-13700K example:
┌─────────────────────────────────────────────┐
│  8 × P-cores  (16 threads via Hyper-Threading)│
│  8 × E-cores  (8 threads, no SMT)             │
│                = 16 cores / 24 threads total  │
└─────────────────────────────────────────────┘

What it looks like: Go back to Figure 2.3 in Section 2. That is a real Intel hybrid chip (a Core i9-13900K, which has 8 P-cores and 16 E-cores). You can see the few big blue P-cores and the many small E-cores sharing one piece of silicon. The pink "Compute Fabric" strip in the middle is the internal road that connects them all.

Windows' scheduler works with Intel's Thread Director technology. Together they try to put demanding, time-sensitive work (like the game or app you are using right now) on P-cores. Background work (like a Windows Update check or a sync client) goes to E-cores.

9.2 AMD Ryzen Architecture

AMD's Ryzen line mostly uses uniform cores: every core is the same design. The cores are grouped in blocks called Core Complex Dies (CCDs). Each CCD holds a cluster of cores (a Core Complex, CCX) that share a slice of L3 cache.

AMD Ryzen 9 example (multi-CCD layout):
┌───────────────────┐   ┌───────────────────┐
│   CCD 0            │   │   CCD 1            │
│  8 cores, shared   │   │  8 cores, shared   │
│  L3 cache          │   │  L3 cache          │
└─────────┬──────────┘   └─────────┬──────────┘
          └──────── Infinity Fabric ────────┘
             (interconnect between CCDs)

All Ryzen cores support SMT (AMD's version of Hyper-Threading). So cores and threads grow together in a simple way (for example, 8 cores / 16 threads).

The Infinity Fabric is the link that joins several CCDs. It matters for speed: two cores on different CCDs talk measurably more slowly than two cores on the same CCD. For some workloads, this makes a difference.

9.3 Performance Differences

"Intel is faster" or "AMD is faster" changes with every product generation. A more useful question is: what kind of work does each design favour?

Practical takeaway: Core count and clock speed on a spec sheet tell only part of the story. The same numbers can give different real-world speed depending on the core design, the cache design, and whether the software actually uses many threads (Section 2.4).

9.4 Power Efficiency Trade-offs

Both companies balance three goals that pull against each other: performance, power use, and heat (which links to TDP and thermal throttling in Section 10).

• Intel's E-cores exist to save power. Running background work on small, low-power cores, instead of waking a big P-core, gives real battery life on laptops.
• AMD's chiplet design means the CCDs are made separately and then joined with Infinity Fabric. It makes manufacturing more efficient and lets AMD add cores cheaply. The cost is extra delay when chiplets talk to each other, and slightly higher idle power from the link itself.
• Both use dynamic frequency scaling (Section 10.2), so a core only draws high power and runs hot when it is doing demanding work. Idle cores automatically drop to a fraction of their peak power.

There is no single "best" design. A chip built for a laptop's battery life makes different trade-offs from one built for a workstation doing 3D rendering all day, even from the same company.

Summary